Currently there are tremendous gaps in our knowledge of who is being prescribed amphetamines, for what indications, and their long term cardiovascular safety. Amphetamines are the 2nd fastest growing class of drugs in the US, driven in part by increasing diagnoses of Attention Deficit Hyperactivity Disorder (ADHD)-which is estimated to affect 1 in 20 adults. Their use for non-FDA approved conditions, like depression, is also expanding. Prior studies indicate that misuse of amphetamines (i.e. abuse) greatly increases the risk of heart attack, stroke, and aortic dissection. While there are physiological reasons to suspect that medical use of prescription amphetamines may increase risk of cardiovascular disease (CVD) outcomes, published epidemiological studies have not answered this question. This national retrospective cohort study of patients treated in the Veterans Affairs (VA) medical system (n=8,000,000; from 2000 to 2008) proposes to: 1) describe the pharmacoepidemiology of prescription amphetamine use, 2) determine which patients are most at risk to progress from medical use to abuse of prescription amphetamines, 3) address the extent to which the medical use of prescription amphetamines is related to CVD outcomes, and 4) if prescription amphetamine use does increase risk of CVD outcomes, describe what subgroups of patients are most at risk. The primary CVD endpoint is a composite outcome of stroke, myocardial infarction, and CVD death. Exposure to prescription amphetamines will be determined using cumulative dose, which accounts for time, dose, and type of amphetamine. Three traditional statistical analytical approaches that seek to adjust for important measured and unmeasured factors that can cause biases in observational studies will be employed: 1) multiple regression, 2) propensity scoring, and 3) instrumental variable estimation. We will also use a novel, more computationally complex and sophisticated statistical approach, called Robust Latent Variable technique (RLV) to adjust for potential biases. RLV is an innovative methodology that has been developed with NIH support over the past ten years which has a number of advantages over propensity scoring and instrumental variable approaches. It incorporates advances for: 1) adjusting for selection bias, 2) minimizing model misspecification, 3) accounting for missing data in the presence of model misspecification, and 4) performing most probable model building to address model uncertainty. These innovative techniques have the potential to broadly impact the analysis of observational studies and advance methods for future comparative effectiveness research. This career award includes a multidisciplinary training plan for obtaining advanced skills in epidemiology, health services research, substance use disorders, and advanced statistical methods. It also includes a multidisciplinary team of mentors with extensive experience in all of these areas with access to national VA data. The results will provide guidance to patients, clinicians, and policymakers regarding the use and safety of prescription amphetamines.